Entering an era of genuine object identification
14 November 2012
HALCON 11, the latest version of MVTec’s image processing library, includes a feature considered to be unique to vision technology which enables an object to be identified by its feature set alone. Suzanne Gill reports.
Julie Busby, technical director at Multipix, a distributor of machine vision technology, explained more about this innovative feature. “Sample-based identification is really one of the highlights of HALCON 11. Essentially, it is a new algorithm that provides a way of identifying a product by sight alone. It can recognise trained objects based on characteristic features such as colour or texture.
“Traditionally, products have been identified by a feature such as a barcode or a data matrix code. To be able to identify a product by its feature set alone is the ‘holy-grail’ of imaging processing,” she enthused.
HALCON is an image processing library containing a powerful development environment. It is used by integrators and developers to create solutions for end-users. It is standard software for machine vision with an integrated development environment (IDE) that is used worldwide and supports a range of operating systems and provides interfaces to hundreds of industrial cameras and frame grabbers, also for standards like GenICam, GigE Vision, and IIDC 1394.
Busby went on to explain more about sample-based identification. “A series of images of an object are presented to the system, from a variety of different perspectives. A large and complex dataset of the object is built up. The system uses this ‘learned’ information to find the object when required.
“The more images of the same product, taken from different perspectives, the more robust the algorithm will become. The algorithm is also able to cope with large variation in the presentation of the product – deformable packaged items that are perspectively challenging can still be distinguished, for example.”
The system can be used in a multitude of applications, where the user is trying to identify what is in front of the camera – it could be used, for example, to identify blister packs, aggregate, different types of food products or deformable packages. It can also identify warped objects or varying perspective views of an object such as a label around a bottle. It is able to cope with different scaling of the product too, as well as variation in terms of rotation and can even cope if only a section of the product can be seen.
A real-world solution
“Sample-based identification offers a more real-world algorithm and also has the ability to greatly reduce development times because of its ability to rapidly create product classifiers or identifiers,” said Busby. “I believe it will open up vision technology for a host of new applications, where the use of an automated vision system, traditionally, would have been to costly or was impossible to create. In other, more traditional areas of automated vision applications, it is certainly helping to speed up development times and making vision solutions a more cost-effective option.”
Another interesting feature of HALCON 11 is the new 3D surface comparison tool, which can identify an expected and measured shape of a 3D object surface. The surface can be reconstructed by any 3D technology available in HALCON like multi-view stereo, sheet of light, or by ready-to-run 3D hardware scanners that are directly supported by HALCON. The 3D surface comparison evaluates the resulting point cloud with the trained object model.
HALCON 11 will now also run on the Mac OS X operating system, entering the Apple world to extend its accessibility
A trail disk of HALCON 11 is available for download from the Mulitpix website: http://multipix.com/
Contact Details and Archive...
Most Viewed Articles...